Working with AWS Clean Rooms ML - AWS Clean Rooms
Working with lookalike models (training data provider)Working with lookalike segments (seed data provider)Next steps

Working with AWS Clean Rooms ML

A lookalike model is a model of a training data provider's data that allows a seed data provider to create a lookalike segment of training data provider's data that most closely resembles their seed data. To create a lookalike model that can be used in a collaboration, you must import your training data, create a lookalike model, configure that lookalike model, and then associate it to a collaboration.

After the training data provider is done creating the ML model, the seed data provider can create and export the seed segment.

Working with lookalike models (training data provider)

Import training data

Before you create a lookalike model, you must specify the AWS Glue table that contains the training data. Clean Rooms ML does not store a copy of this data, just metadata that allows it to access the data.

To import training data in AWS Clean Rooms

  1. Sign in to the AWS Management Console and open the AWS Clean Rooms console with your AWS account (if you have not yet done so).

  2. In the left navigation pane, choose ML Modeling.

  3. On the Training datasets tab, choose Create training dataset.

  4. Enter a Name and optional Description.

  5. For Data source, choose your AWS Glue table:

    1. Choose the Database that you want to configure from the dropdown list.

    2. Choose the Training data source by selecting the Database and Table that you want to configure from the dropdown lists.

    Note

    To verify that this is the correct table, do either one of the following:

    • Choose View in AWS Glue.

    • Turn on View schema to view the schema.

  6. For Training details, choose the User identifier column, Item identifier column, and Timestamp column from your data. The training data must contain these three columns. You can also select any other columns that you want to include in the training data.

    The data in the Timestamp column must be in the Unix epoch time in seconds format.

  7. In Service access, you must specify a service role that can access your data and provide a KMS key if your data is encrypted. Choose Create and use a new service role and Clean Rooms ML will automatically create a service role and add the necessary permissions policy. Choose Use an existing service role and enter it in the Service role name field if you have a specific service role that you want to use.

    If your data is encrypted, enter your KMS key in the AWS KMS key field, or click Create an AWS KMS key to generate a new KMS key.

  8. If you want to enable Tags for the training dataset, choose Add new tag and then enter the Key and Value pair.

  9. Choose Create training dataset.

For the corresponding API action, see CreateTrainingDataset.

Create a lookalike model

After you have created a training dataset, you are ready to create a lookalike model. You can create many lookalike models from a single training dataset.

You must create a default database in your AWS Glue Data Catalog or include the glue:createDatabase permission in the provided role.

To create a lookalike model in AWS Clean Rooms

  1. Sign in to the AWS Management Console and open the AWS Clean Rooms console with your AWS account (if you have not yet done so).

  2. In the left navigation pane, choose ML Modeling.

  3. On the Lookalike models tab, choose Create lookalike model.

  4. For Create lookalike model, for Lookalike model details:

    1. Enter a Name and optional Description.

    2. Choose the Training dataset that you want to model from the dropdown list.

    3. Enter an optional Training window.

  5. If you want to enable custom encryption settings for the lookalike model, choose Customize encryption settings and then enter the KMS key.

  6. If you want to enable Tags for the lookalike model, choose Add new tag and then enter the Key and Value pair.

  7. Choose Create lookalike model.

For the corresponding API action, see CreateAudienceModel.

Configure a lookalike model

After you have created a lookalike model, you are ready to configure it for use in a collaboration. You can create multiple configured lookalike models from a single lookalike model.

To configure a lookalike model in AWS Clean Rooms

  1. Sign in to the AWS Management Console and open the AWS Clean Rooms console with your AWS account (if you have not yet done so).

  2. In the left navigation pane, choose ML Modeling.

  3. On the Configured lookalike models tab, choose Configure lookalike model.

  4. For Configure lookalike model, for Configured lookalike model details:

    1. Enter a Name and optional Description.

    2. Choose the Lookalike model that you want to configure from the dropdown list.

    3. Choose the Minimum matching seed size that you want. This is the minimum number of users in the seed data provider's data that overlap with users in the training data. This value must be greater than 0.

  5. For Metrics to share with other members, choose whether you want the seed data provider in your collaboration to receive model metrics, including relevance scores.

  6. For Lookalike segment destination location, enter the Amazon S3 bucket where lookalike segment is exported. This bucket must be located in the same region as your other resources.

  7. For Service access, choose the Existing service role name that will be used to access this table.

  8. Choose Configure Lookalike Model.

  9. If you want to enable Tags for the configured table resource, choose Add new tag and then enter the Key and Value pair.

For the corresponding API action, see CreateConfiguredAudienceModel.

Associate a configured lookalike model

After you have configured a lookalike model, you can associate it to a collaboration.

To associate a configured lookalike model in AWS Clean Rooms

  1. Sign in to the AWS Management Console and open the AWS Clean Rooms console with your AWS account (if you have not yet done so).

  2. In the left navigation pane, choose Collaborations.

  3. On the With active membership tab, choose a collaboration.

  4. On the ML Modeling tab, choose Associate lookalike model.

  5. For Associate configured lookalike model, for Associate lookalike model details:

    1. Enter a Name for the associated configured audience model.

    2. Enter a Description of the table.

      The description helps differentiate between other associated configured audience models with similar names.

  6. For Configured lookalike model, choose a configured lookalike model from the dropdown list.

  7. Choose Associate.

For the corresponding API action, see CreateConfiguredAudienceModelAssociation.

Update a configured lookalike model

After you have associated a configured a lookalike model, you can update it to change information such as the name, metrics to share, or output Amazon S3 location.

To update an associated configured lookalike model in AWS Clean Rooms

  1. Sign in to the AWS Management Console and open the AWS Clean Rooms console with your AWS account (if you have not yet done so).

  2. In the left navigation pane, choose ML modeling.

  3. On the Configured lookalike models tab, choose a configured lookalike model and select Edit.

  4. For Configure lookalike model, for Configured lookalike model details:

    1. Choose the Lookalike model that you want configured from the dropdown list.

    2. Choose the Minimum matching seed size that you want. This is the minimum number of users in the seed data provider's data that overlap with users in the training data. This value must be greater than 0.

  5. For Metrics to share with other members, choose whether you want the seed data provider in your collaboration to receive model metrics, including relevance scores.

  6. For Lookalike segment destination location, enter the Amazon S3 bucket where lookalike segment is exported. This bucket must be located in the same region as your other resources.

  7. For Service access, choose the Existing service role name that will be used to access this table.

  8. For Advanced bin size configuration, choose how you want to configure the audience bin sizes.

  9. Choose Save changes.

For the corresponding API action, see UpdateConfiguredAudienceModel.

Working with lookalike segments (seed data provider)

Create a lookalike segment

A lookalike segment is a subset of the training data that most closely resembles the seed data.

To create a lookalike segment in AWS Clean Rooms

  1. Sign in to the AWS Management Console and open the AWS Clean Rooms console with your AWS account (if you have not yet done so).

  2. In the left navigation pane, choose Collaborations.

  3. On the With active membership tab, choose a collaboration.

  4. On the ML Modeling tab, choose Create lookalike segment.

  5. For Create lookalike segment, for Lookalike segment details enter a Name and optional Description.

  6. For Seed profiles, choose the Amazon S3 input source where your seed data is stored.

  7. For Service access, choose the Existing service role name that will be used to access this table.

  8. If you want to enable Tags for the training dataset, choose Add new tag and then enter the Key and Value pair.

  9. Choose Create lookalike segment.

For the corresponding API action, see StartAudienceGenerationJob.

Export a lookalike segment

After you have created a lookalike segment, you can export that data to an Amazon S3 bucket.

To export a lookalike segment in AWS Clean Rooms

  1. Sign in to the AWS Management Console and open the AWS Clean Rooms console with your AWS account (if you have not yet done so).

  2. In the left navigation pane, choose Collaborations.

  3. On the With active membership tab, choose a collaboration.

  4. On the ML Modeling tab, select a lookalike segment and choose Export.

  5. For Export lookalike model, for Export lookalike model details enter a Name and optional Description.

  6. For Segment size, choose the size you want for the exported segment.

  7. Choose Export.

For the corresponding API action, see StartAudienceExportJob.

Next steps

Now that you have created a lookalike model and exported a seed segment, you are ready to: